Culling For Context (insider notes)

Without going into details explaining some of the propriotary code involved in this writeup, I will share the basic overview.  This document was writtin in 2007. – lkw

Intro
Context resolution first depends upon our ability to cull a clump, a phrase chain, and a neuron. Culling occurs at different stages of a parse. The timing of the cull changes how the cull operates. There are pre-parse culls that represent the words and there immediate associations. This mimics the brains ability to fire off immediate associations based on the word, not necessarily the intended concept of the word or phrase. Those follow later after they are resolved. A demonstratable example of this “instinctual cull” occurs with phoneme sets, as when a word like “sextuplet” is used. Many people may first hear “sex” and key off it. In doing so, they fire off immediate associations. We do not handle phonemes right now, but when we do this will be another type of pre-parse cull. In these cases, firing will occur, but at a very low level.

Other culls happen after the parse is complete. At this stage all the phrases and neurons that could be resolved have been assigned their known values. These resolved values are culled, fired, and placed into a pool (g_chain). Everything culled from a specific Nid, Cid, or Role has a path back to the originating culler (through linked g_links). In other words, if the instance “Fido” caused dog, animal and lifeform to be in the context pool, those 3 specific “g_links” would have paths back to the original instance that got them in the pool. This is an important factor when considering a g_chain (pool) and its links. It operates slightly different than the traditional context pool because it needs the ability to hold duplicate entries. This allows the concept of dog to be in the pool 3 times if there are 3 unique dogs.   In other words, the concept neuron ‘dog’ has been fired 3 times.    This structure allows for us to know why the neuron is firing.

The aforementioned culls get data into pools. This data is used to resolve phrases, cids, and nids during parse time. Phrase links are the core constituents of a parse. A sentence is made of phrases, and these phrases are referencing:

  1. something in context
  2. something in memory, or
  3. something new we are not familiar with.

At parse time, whenever a phrase is parsed, we attempt to resolve it by looking to the g_chain (pool) for relevant data. We look in memory for something out-of-context.

Recap

Functions of culling:
1. Cull to Chain/Link (Formerly a Pool/Table… Dupes necessary)
2. Cull for the purpose of firing the neurons. (dupes not necessary, firing is boosted)
3. Do both. (Perhaps in a future system they are one in the same)

CULLING IMPORTANCE
In Context.h I created and used an enum called cp_timing. It directs my culls on how to cull

/*** Context cull type used in weighting firing levels ***
We will call culls at different stages of the parse to help with context and splitter resolution. The type helps determine the firing/weighting level. Before resolution, just addressing the neuron should fire it to some level. A stronger firing will occur when we resolve which splitter we are going with. */

enumcp_timing {CPCT_PRE_PARSE=1, CPCT_SPLIT_RESOLUTION, CPCT_NEURON_RESOLUTION, CPCT_CLUMP_INST_BUILDOUT, CPCT_INST, CPCT_CC_TOPIC_BUILD, CPCT_FINAL_FIRE, CPCT_CONJECT, CPCT_THES, CPCT_CMD};

All of these culls will need to be re-written in light of our change from tables to chains. It is absolutely critical that this is handled correctly. The resolution of phrases and concepts in context depends upon the correct concepts being put in the context pools. They are my indexes into various cids and nids. In order to attempt to resolve the phrase “the dog” in context, we will look for the concept of “dog” in the pool. From there we will track it through the g_links to the instance that put it into the pool (if any).

Resolution Challenges (a side note)
It is important to note, when resolving context, we don’t always know what we are looking for. Even “the chicken” doesn’t have to be a traditional instance. Generally the determiner “the” clues us in to focus on instances, but it isn’t a guarantee.

Proverbial Instances
The sentence could be, “the chicken that bites is a chicken not to be trusted.” I refer to these fake instances as “Proverbial Instances”. The phrase “the chicken” is masquerading as an instance, when in fact it is still a class definition not too different than the class definitions:

Chickens that bite are not to be trusted.”

Biting chickens should not be trusted.” (‘Biting chickens’ can be ambiguous here! Those that bite, or the act of biting a chicken!)

Don’t trust chickens that bite.”

Proverbial Instances capture “words of wisdom” statements (he who ….) and they capture parable like statements as in, “the man that buried his money…”

As is the case with all faux instances, the challenge is isolating the cases and determining where to store them! Should we reference these in the class definition, or should they be treated as special Proverbial Instances?

Cultural Instances
Another masquerading instance case is the “Cultural Instance”. Why did the chicken cross the road? This is not an instance per say, it is a “Cultural Instance”; the same chicken everyone talks about when talking about “the chicken” that crossed the road. It isn’t a real chicken, it is a proverbial chicken used in a “thought experiment” or as a cultural example.

Side Tangent: Everyone has their own opinion why this chicken journeyed across the road. Perhaps Plato would say it was for the greater good. Thomas Visel might add it was because by crossing the road the chicken saved 2 bits of memory! The woman who cooks steak in two pans would say the chicken crossed the road because that is what the chicken always does! Schrödinger might think it was on both sides of the road, crossing, and not crossing, all at the same time! Karl Marx would contend it was a historical inevitability. Oliver North may think it was because National Security was at stake, and one might argue that Einstein would propose that whether or not the chicken actually crossed the road depends upon your frame of reference!

Clonable Instance
Finally, there is the faux instance that itself that can be instantiated. In this case, it at first appears to be an instance, but it truly is a class. At first glance, “The Bible” is an instance of a book, but upon inspection it really is a “Class” of books. (child of book) “The Bible” can be instantiated because I can own a copy, and so can you! This seems to be a unique property of things that can be duplicated in exact form. I am calling this the “Clonable Instance”. We have unique instances, but all of the instances can be referred to. The same can be said of a specific book by a specific person. I own a particular copy of Ray Kurzweil’s “The Singularity is Near”. My copy has hand written notes (by me) and specific pages are bent. Unique creases, nicks and dings, as well as fade marks are on my copy. This is a true instance of Ray’s book, and it IS NOT the same as the clonable instance of a book referred to when someone says, “Ray Kurzweil’s new book The Singularity Is Near just hit the shelves. That book is amazing!

Notes on Intent
In many cases, faux instances are used to drive home a point in a less offensive way. This can be a bragging point about oneself or an insulting point about another. They are also used for humor or to illuminate a greater truth. It seems like faux instances offer a unique method of elucidating a proposed or believed truth. Those who are able will see it. (Whether it is in fact true or not) Those who are not able to see it, will not be aware of the hidden layer. We may be in a discussion where both parties are aware that someone is a believer in “aliens”. Instead of one of us saying, “you are crazy for believing in stuff like that!”, we may opt say something like, “The man who believes in unproven and mythical things is a man who is of greater faith than I”. In a round-about way we have called someone something by defining a class that they obviously are a member of! This information will be captured at higher “meta levels” of our awareness pool. By asking the internal question of, “What does this mean?” for every input received, we are able to draw the conclusion (new awareness) that the speaker is providing class defining information of a class one or both of us is clearly apart of.

Conclusion
Culling must be done right, for it sets up and enables proper resolution.

 

Advertisements
Previous Post
Leave a comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: